Xiaodong-Bran / MLFcGAN

Xiaodong-Bran/MLFcGAN-Multilevel-Feature-Fusion-Based-Conditional-GAN-for-Underwater-Image-Color-Correction

PyTorch implementation of MLFcGAN: Multilevel Feature Fusion-Based Conditional GAN for Underwater Image Color Correction .

Based on pix2pix by Phillip Isola et al.

Prerequisites

• Linux
• Python, Numpy, PIL
• pytorch 1.2.0
• torchvision 0.4.0

Getting Started

• Clone this repo:

  git clone [e-mail]:Xiaodong-Bran/MLFcGAN.git

  cd MLFcGAN

• download the pre-trained model: [google-dirve] (https://drive.google.com/open?id=1OREuAj6DplD0-ipQ3s37aZ6j9Q5kXvtO)

• Prepare the underwater image dataset.The structure of the image folders should follows:

name_of_dataset/
└── source2target
    ├── test
    │   ├── source
    │   └── target
    └── train
        ├── source
        └── target

• Modifiy the test_img_folder and test_output_path in test.sh

• To test the model, please run:

  sh test.sh

• To train the model, please run:

  sh train.sh

Acknowledgments

This code is inspired by pix2pix.

Highly recommend the more sophisticated and organized code pytorch-CycleGAN-and-pix2pix by Jun-Yan Zhu.